1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PARAVIRT_H
3 #define _ASM_X86_PARAVIRT_H
4 /* Various instructions on x86 need to be replaced for
5 * para-virtualization: those hooks are defined here. */
8 #include <asm/pgtable_types.h>
10 #include <asm/nospec-branch.h>
12 #include <asm/paravirt_types.h>
15 #include <linux/bug.h>
16 #include <linux/types.h>
17 #include <linux/cpumask.h>
18 #include <linux/static_call_types.h>
19 #include <asm/frame.h>
21 u64 dummy_steal_clock(int cpu);
22 u64 dummy_sched_clock(void);
24 DECLARE_STATIC_CALL(pv_steal_clock, dummy_steal_clock);
25 DECLARE_STATIC_CALL(pv_sched_clock, dummy_sched_clock);
27 void paravirt_set_sched_clock(u64 (*func)(void));
29 static inline u64 paravirt_sched_clock(void)
31 return static_call(pv_sched_clock)();
35 extern struct static_key paravirt_steal_enabled;
36 extern struct static_key paravirt_steal_rq_enabled;
38 __visible void __native_queued_spin_unlock(struct qspinlock *lock);
39 bool pv_is_native_spin_unlock(void);
40 __visible bool __native_vcpu_is_preempted(long cpu);
41 bool pv_is_native_vcpu_is_preempted(void);
43 static inline u64 paravirt_steal_clock(int cpu)
45 return static_call(pv_steal_clock)(cpu);
48 #ifdef CONFIG_PARAVIRT_SPINLOCKS
49 void __init paravirt_set_cap(void);
52 /* The paravirtualized I/O functions */
53 static inline void slow_down_io(void)
55 pv_ops.cpu.io_delay();
57 pv_ops.cpu.io_delay();
58 pv_ops.cpu.io_delay();
59 pv_ops.cpu.io_delay();
63 void native_flush_tlb_local(void);
64 void native_flush_tlb_global(void);
65 void native_flush_tlb_one_user(unsigned long addr);
66 void native_flush_tlb_multi(const struct cpumask *cpumask,
67 const struct flush_tlb_info *info);
69 static inline void __flush_tlb_local(void)
71 PVOP_VCALL0(mmu.flush_tlb_user);
74 static inline void __flush_tlb_global(void)
76 PVOP_VCALL0(mmu.flush_tlb_kernel);
79 static inline void __flush_tlb_one_user(unsigned long addr)
81 PVOP_VCALL1(mmu.flush_tlb_one_user, addr);
84 static inline void __flush_tlb_multi(const struct cpumask *cpumask,
85 const struct flush_tlb_info *info)
87 PVOP_VCALL2(mmu.flush_tlb_multi, cpumask, info);
90 static inline void paravirt_tlb_remove_table(struct mmu_gather *tlb, void *table)
92 PVOP_VCALL2(mmu.tlb_remove_table, tlb, table);
95 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
97 PVOP_VCALL1(mmu.exit_mmap, mm);
100 static inline void notify_page_enc_status_changed(unsigned long pfn,
101 int npages, bool enc)
103 PVOP_VCALL3(mmu.notify_page_enc_status_changed, pfn, npages, enc);
106 #ifdef CONFIG_PARAVIRT_XXL
107 static inline void load_sp0(unsigned long sp0)
109 PVOP_VCALL1(cpu.load_sp0, sp0);
112 /* The paravirtualized CPUID instruction. */
113 static inline void __cpuid(unsigned int *eax, unsigned int *ebx,
114 unsigned int *ecx, unsigned int *edx)
116 PVOP_VCALL4(cpu.cpuid, eax, ebx, ecx, edx);
120 * These special macros can be used to get or set a debugging register
122 static inline unsigned long paravirt_get_debugreg(int reg)
124 return PVOP_CALL1(unsigned long, cpu.get_debugreg, reg);
126 #define get_debugreg(var, reg) var = paravirt_get_debugreg(reg)
127 static inline void set_debugreg(unsigned long val, int reg)
129 PVOP_VCALL2(cpu.set_debugreg, reg, val);
132 static inline unsigned long read_cr0(void)
134 return PVOP_CALL0(unsigned long, cpu.read_cr0);
137 static inline void write_cr0(unsigned long x)
139 PVOP_VCALL1(cpu.write_cr0, x);
142 static inline unsigned long read_cr2(void)
144 return PVOP_ALT_CALLEE0(unsigned long, mmu.read_cr2,
146 ALT_NOT(X86_FEATURE_XENPV));
149 static inline void write_cr2(unsigned long x)
151 PVOP_VCALL1(mmu.write_cr2, x);
154 static inline unsigned long __read_cr3(void)
156 return PVOP_ALT_CALL0(unsigned long, mmu.read_cr3,
157 "mov %%cr3, %%rax;", ALT_NOT(X86_FEATURE_XENPV));
160 static inline void write_cr3(unsigned long x)
162 PVOP_ALT_VCALL1(mmu.write_cr3, x,
163 "mov %%rdi, %%cr3", ALT_NOT(X86_FEATURE_XENPV));
166 static inline void __write_cr4(unsigned long x)
168 PVOP_VCALL1(cpu.write_cr4, x);
171 static inline void arch_safe_halt(void)
173 PVOP_VCALL0(irq.safe_halt);
176 static inline void halt(void)
178 PVOP_VCALL0(irq.halt);
181 static inline void wbinvd(void)
183 PVOP_ALT_VCALL0(cpu.wbinvd, "wbinvd", ALT_NOT(X86_FEATURE_XENPV));
186 static inline u64 paravirt_read_msr(unsigned msr)
188 return PVOP_CALL1(u64, cpu.read_msr, msr);
191 static inline void paravirt_write_msr(unsigned msr,
192 unsigned low, unsigned high)
194 PVOP_VCALL3(cpu.write_msr, msr, low, high);
197 static inline u64 paravirt_read_msr_safe(unsigned msr, int *err)
199 return PVOP_CALL2(u64, cpu.read_msr_safe, msr, err);
202 static inline int paravirt_write_msr_safe(unsigned msr,
203 unsigned low, unsigned high)
205 return PVOP_CALL3(int, cpu.write_msr_safe, msr, low, high);
208 #define rdmsr(msr, val1, val2) \
210 u64 _l = paravirt_read_msr(msr); \
215 #define wrmsr(msr, val1, val2) \
217 paravirt_write_msr(msr, val1, val2); \
220 #define rdmsrl(msr, val) \
222 val = paravirt_read_msr(msr); \
225 static inline void wrmsrl(unsigned msr, u64 val)
227 wrmsr(msr, (u32)val, (u32)(val>>32));
230 #define wrmsr_safe(msr, a, b) paravirt_write_msr_safe(msr, a, b)
232 /* rdmsr with exception handling */
233 #define rdmsr_safe(msr, a, b) \
236 u64 _l = paravirt_read_msr_safe(msr, &_err); \
242 static inline int rdmsrl_safe(unsigned msr, unsigned long long *p)
246 *p = paravirt_read_msr_safe(msr, &err);
250 static inline unsigned long long paravirt_read_pmc(int counter)
252 return PVOP_CALL1(u64, cpu.read_pmc, counter);
255 #define rdpmc(counter, low, high) \
257 u64 _l = paravirt_read_pmc(counter); \
262 #define rdpmcl(counter, val) ((val) = paravirt_read_pmc(counter))
264 static inline void paravirt_alloc_ldt(struct desc_struct *ldt, unsigned entries)
266 PVOP_VCALL2(cpu.alloc_ldt, ldt, entries);
269 static inline void paravirt_free_ldt(struct desc_struct *ldt, unsigned entries)
271 PVOP_VCALL2(cpu.free_ldt, ldt, entries);
274 static inline void load_TR_desc(void)
276 PVOP_VCALL0(cpu.load_tr_desc);
278 static inline void load_gdt(const struct desc_ptr *dtr)
280 PVOP_VCALL1(cpu.load_gdt, dtr);
282 static inline void load_idt(const struct desc_ptr *dtr)
284 PVOP_VCALL1(cpu.load_idt, dtr);
286 static inline void set_ldt(const void *addr, unsigned entries)
288 PVOP_VCALL2(cpu.set_ldt, addr, entries);
290 static inline unsigned long paravirt_store_tr(void)
292 return PVOP_CALL0(unsigned long, cpu.store_tr);
295 #define store_tr(tr) ((tr) = paravirt_store_tr())
296 static inline void load_TLS(struct thread_struct *t, unsigned cpu)
298 PVOP_VCALL2(cpu.load_tls, t, cpu);
301 static inline void load_gs_index(unsigned int gs)
303 PVOP_VCALL1(cpu.load_gs_index, gs);
306 static inline void write_ldt_entry(struct desc_struct *dt, int entry,
309 PVOP_VCALL3(cpu.write_ldt_entry, dt, entry, desc);
312 static inline void write_gdt_entry(struct desc_struct *dt, int entry,
313 void *desc, int type)
315 PVOP_VCALL4(cpu.write_gdt_entry, dt, entry, desc, type);
318 static inline void write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
320 PVOP_VCALL3(cpu.write_idt_entry, dt, entry, g);
323 #ifdef CONFIG_X86_IOPL_IOPERM
324 static inline void tss_invalidate_io_bitmap(void)
326 PVOP_VCALL0(cpu.invalidate_io_bitmap);
329 static inline void tss_update_io_bitmap(void)
331 PVOP_VCALL0(cpu.update_io_bitmap);
335 static inline void paravirt_activate_mm(struct mm_struct *prev,
336 struct mm_struct *next)
338 PVOP_VCALL2(mmu.activate_mm, prev, next);
341 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
342 struct mm_struct *mm)
344 PVOP_VCALL2(mmu.dup_mmap, oldmm, mm);
347 static inline int paravirt_pgd_alloc(struct mm_struct *mm)
349 return PVOP_CALL1(int, mmu.pgd_alloc, mm);
352 static inline void paravirt_pgd_free(struct mm_struct *mm, pgd_t *pgd)
354 PVOP_VCALL2(mmu.pgd_free, mm, pgd);
357 static inline void paravirt_alloc_pte(struct mm_struct *mm, unsigned long pfn)
359 PVOP_VCALL2(mmu.alloc_pte, mm, pfn);
361 static inline void paravirt_release_pte(unsigned long pfn)
363 PVOP_VCALL1(mmu.release_pte, pfn);
366 static inline void paravirt_alloc_pmd(struct mm_struct *mm, unsigned long pfn)
368 PVOP_VCALL2(mmu.alloc_pmd, mm, pfn);
371 static inline void paravirt_release_pmd(unsigned long pfn)
373 PVOP_VCALL1(mmu.release_pmd, pfn);
376 static inline void paravirt_alloc_pud(struct mm_struct *mm, unsigned long pfn)
378 PVOP_VCALL2(mmu.alloc_pud, mm, pfn);
380 static inline void paravirt_release_pud(unsigned long pfn)
382 PVOP_VCALL1(mmu.release_pud, pfn);
385 static inline void paravirt_alloc_p4d(struct mm_struct *mm, unsigned long pfn)
387 PVOP_VCALL2(mmu.alloc_p4d, mm, pfn);
390 static inline void paravirt_release_p4d(unsigned long pfn)
392 PVOP_VCALL1(mmu.release_p4d, pfn);
395 static inline pte_t __pte(pteval_t val)
397 return (pte_t) { PVOP_ALT_CALLEE1(pteval_t, mmu.make_pte, val,
399 ALT_NOT(X86_FEATURE_XENPV)) };
402 static inline pteval_t pte_val(pte_t pte)
404 return PVOP_ALT_CALLEE1(pteval_t, mmu.pte_val, pte.pte,
405 "mov %%rdi, %%rax", ALT_NOT(X86_FEATURE_XENPV));
408 static inline pgd_t __pgd(pgdval_t val)
410 return (pgd_t) { PVOP_ALT_CALLEE1(pgdval_t, mmu.make_pgd, val,
412 ALT_NOT(X86_FEATURE_XENPV)) };
415 static inline pgdval_t pgd_val(pgd_t pgd)
417 return PVOP_ALT_CALLEE1(pgdval_t, mmu.pgd_val, pgd.pgd,
418 "mov %%rdi, %%rax", ALT_NOT(X86_FEATURE_XENPV));
421 #define __HAVE_ARCH_PTEP_MODIFY_PROT_TRANSACTION
422 static inline pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr,
427 ret = PVOP_CALL3(pteval_t, mmu.ptep_modify_prot_start, vma, addr, ptep);
429 return (pte_t) { .pte = ret };
432 static inline void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr,
433 pte_t *ptep, pte_t old_pte, pte_t pte)
436 PVOP_VCALL4(mmu.ptep_modify_prot_commit, vma, addr, ptep, pte.pte);
439 static inline void set_pte(pte_t *ptep, pte_t pte)
441 PVOP_VCALL2(mmu.set_pte, ptep, pte.pte);
444 static inline void set_pmd(pmd_t *pmdp, pmd_t pmd)
446 PVOP_VCALL2(mmu.set_pmd, pmdp, native_pmd_val(pmd));
449 static inline pmd_t __pmd(pmdval_t val)
451 return (pmd_t) { PVOP_ALT_CALLEE1(pmdval_t, mmu.make_pmd, val,
453 ALT_NOT(X86_FEATURE_XENPV)) };
456 static inline pmdval_t pmd_val(pmd_t pmd)
458 return PVOP_ALT_CALLEE1(pmdval_t, mmu.pmd_val, pmd.pmd,
459 "mov %%rdi, %%rax", ALT_NOT(X86_FEATURE_XENPV));
462 static inline void set_pud(pud_t *pudp, pud_t pud)
464 PVOP_VCALL2(mmu.set_pud, pudp, native_pud_val(pud));
467 static inline pud_t __pud(pudval_t val)
471 ret = PVOP_ALT_CALLEE1(pudval_t, mmu.make_pud, val,
472 "mov %%rdi, %%rax", ALT_NOT(X86_FEATURE_XENPV));
474 return (pud_t) { ret };
477 static inline pudval_t pud_val(pud_t pud)
479 return PVOP_ALT_CALLEE1(pudval_t, mmu.pud_val, pud.pud,
480 "mov %%rdi, %%rax", ALT_NOT(X86_FEATURE_XENPV));
483 static inline void pud_clear(pud_t *pudp)
485 set_pud(pudp, native_make_pud(0));
488 static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
490 p4dval_t val = native_p4d_val(p4d);
492 PVOP_VCALL2(mmu.set_p4d, p4dp, val);
495 #if CONFIG_PGTABLE_LEVELS >= 5
497 static inline p4d_t __p4d(p4dval_t val)
499 p4dval_t ret = PVOP_ALT_CALLEE1(p4dval_t, mmu.make_p4d, val,
501 ALT_NOT(X86_FEATURE_XENPV));
503 return (p4d_t) { ret };
506 static inline p4dval_t p4d_val(p4d_t p4d)
508 return PVOP_ALT_CALLEE1(p4dval_t, mmu.p4d_val, p4d.p4d,
509 "mov %%rdi, %%rax", ALT_NOT(X86_FEATURE_XENPV));
512 static inline void __set_pgd(pgd_t *pgdp, pgd_t pgd)
514 PVOP_VCALL2(mmu.set_pgd, pgdp, native_pgd_val(pgd));
517 #define set_pgd(pgdp, pgdval) do { \
518 if (pgtable_l5_enabled()) \
519 __set_pgd(pgdp, pgdval); \
521 set_p4d((p4d_t *)(pgdp), (p4d_t) { (pgdval).pgd }); \
524 #define pgd_clear(pgdp) do { \
525 if (pgtable_l5_enabled()) \
526 set_pgd(pgdp, native_make_pgd(0)); \
529 #endif /* CONFIG_PGTABLE_LEVELS == 5 */
531 static inline void p4d_clear(p4d_t *p4dp)
533 set_p4d(p4dp, native_make_p4d(0));
536 static inline void set_pte_atomic(pte_t *ptep, pte_t pte)
541 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
544 set_pte(ptep, native_make_pte(0));
547 static inline void pmd_clear(pmd_t *pmdp)
549 set_pmd(pmdp, native_make_pmd(0));
552 #define __HAVE_ARCH_START_CONTEXT_SWITCH
553 static inline void arch_start_context_switch(struct task_struct *prev)
555 PVOP_VCALL1(cpu.start_context_switch, prev);
558 static inline void arch_end_context_switch(struct task_struct *next)
560 PVOP_VCALL1(cpu.end_context_switch, next);
563 #define __HAVE_ARCH_ENTER_LAZY_MMU_MODE
564 static inline void arch_enter_lazy_mmu_mode(void)
566 PVOP_VCALL0(mmu.lazy_mode.enter);
569 static inline void arch_leave_lazy_mmu_mode(void)
571 PVOP_VCALL0(mmu.lazy_mode.leave);
574 static inline void arch_flush_lazy_mmu_mode(void)
576 PVOP_VCALL0(mmu.lazy_mode.flush);
579 static inline void __set_fixmap(unsigned /* enum fixed_addresses */ idx,
580 phys_addr_t phys, pgprot_t flags)
582 pv_ops.mmu.set_fixmap(idx, phys, flags);
586 #if defined(CONFIG_SMP) && defined(CONFIG_PARAVIRT_SPINLOCKS)
588 static __always_inline void pv_queued_spin_lock_slowpath(struct qspinlock *lock,
591 PVOP_VCALL2(lock.queued_spin_lock_slowpath, lock, val);
594 static __always_inline void pv_queued_spin_unlock(struct qspinlock *lock)
596 PVOP_ALT_VCALLEE1(lock.queued_spin_unlock, lock,
597 "movb $0, (%%" _ASM_ARG1 ");",
598 ALT_NOT(X86_FEATURE_PVUNLOCK));
601 static __always_inline void pv_wait(u8 *ptr, u8 val)
603 PVOP_VCALL2(lock.wait, ptr, val);
606 static __always_inline void pv_kick(int cpu)
608 PVOP_VCALL1(lock.kick, cpu);
611 static __always_inline bool pv_vcpu_is_preempted(long cpu)
613 return PVOP_ALT_CALLEE1(bool, lock.vcpu_is_preempted, cpu,
614 "xor %%" _ASM_AX ", %%" _ASM_AX ";",
615 ALT_NOT(X86_FEATURE_VCPUPREEMPT));
618 void __raw_callee_save___native_queued_spin_unlock(struct qspinlock *lock);
619 bool __raw_callee_save___native_vcpu_is_preempted(long cpu);
621 #endif /* SMP && PARAVIRT_SPINLOCKS */
624 /* save and restore all caller-save registers, except return value */
625 #define PV_SAVE_ALL_CALLER_REGS "pushl %ecx;"
626 #define PV_RESTORE_ALL_CALLER_REGS "popl %ecx;"
628 /* save and restore all caller-save registers, except return value */
629 #define PV_SAVE_ALL_CALLER_REGS \
638 #define PV_RESTORE_ALL_CALLER_REGS \
650 * Generate a thunk around a function which saves all caller-save
651 * registers except for the return value. This allows C functions to
652 * be called from assembler code where fewer than normal registers are
653 * available. It may also help code generation around calls from C
654 * code if the common case doesn't use many registers.
656 * When a callee is wrapped in a thunk, the caller can assume that all
657 * arg regs and all scratch registers are preserved across the
658 * call. The return value in rax/eax will not be saved, even for void
661 #define PV_THUNK_NAME(func) "__raw_callee_save_" #func
662 #define PV_CALLEE_SAVE_REGS_THUNK(func) \
663 extern typeof(func) __raw_callee_save_##func; \
665 asm(".pushsection .text;" \
666 ".globl " PV_THUNK_NAME(func) ";" \
667 ".type " PV_THUNK_NAME(func) ", @function;" \
668 PV_THUNK_NAME(func) ":" \
670 PV_SAVE_ALL_CALLER_REGS \
672 PV_RESTORE_ALL_CALLER_REGS \
675 ".size " PV_THUNK_NAME(func) ", .-" PV_THUNK_NAME(func) ";" \
678 /* Get a reference to a callee-save function */
679 #define PV_CALLEE_SAVE(func) \
680 ((struct paravirt_callee_save) { __raw_callee_save_##func })
682 /* Promise that "func" already uses the right calling convention */
683 #define __PV_IS_CALLEE_SAVE(func) \
684 ((struct paravirt_callee_save) { func })
686 #ifdef CONFIG_PARAVIRT_XXL
687 static inline notrace unsigned long arch_local_save_flags(void)
689 return PVOP_ALT_CALLEE0(unsigned long, irq.save_fl, "pushf; pop %%rax;",
690 ALT_NOT(X86_FEATURE_XENPV));
693 static inline notrace void arch_local_irq_disable(void)
695 PVOP_ALT_VCALLEE0(irq.irq_disable, "cli;", ALT_NOT(X86_FEATURE_XENPV));
698 static inline notrace void arch_local_irq_enable(void)
700 PVOP_ALT_VCALLEE0(irq.irq_enable, "sti;", ALT_NOT(X86_FEATURE_XENPV));
703 static inline notrace unsigned long arch_local_irq_save(void)
707 f = arch_local_save_flags();
708 arch_local_irq_disable();
714 /* Make sure as little as possible of this mess escapes. */
729 extern void default_banner(void);
731 #else /* __ASSEMBLY__ */
733 #define _PVSITE(ptype, ops, word, algn) \
737 .pushsection .parainstructions,"a"; \
746 #ifdef CONFIG_PARAVIRT_XXL
748 #define PARA_PATCH(off) ((off) / 8)
749 #define PARA_SITE(ptype, ops) _PVSITE(ptype, ops, .quad, 8)
750 #define PARA_INDIRECT(addr) *addr(%rip)
752 #define INTERRUPT_RETURN \
753 ANNOTATE_RETPOLINE_SAFE; \
754 ALTERNATIVE_TERNARY("jmp *paravirt_iret(%rip);", \
755 X86_FEATURE_XENPV, "jmp xen_iret;", "jmp native_iret;")
757 #ifdef CONFIG_DEBUG_ENTRY
758 .macro PARA_IRQ_save_fl
759 PARA_SITE(PARA_PATCH(PV_IRQ_save_fl),
760 ANNOTATE_RETPOLINE_SAFE;
761 call PARA_INDIRECT(pv_ops+PV_IRQ_save_fl);)
764 #define SAVE_FLAGS ALTERNATIVE "PARA_IRQ_save_fl;", "pushf; pop %rax;", \
765 ALT_NOT(X86_FEATURE_XENPV)
767 #endif /* CONFIG_PARAVIRT_XXL */
768 #endif /* CONFIG_X86_64 */
770 #endif /* __ASSEMBLY__ */
771 #else /* CONFIG_PARAVIRT */
772 # define default_banner x86_init_noop
773 #endif /* !CONFIG_PARAVIRT */
776 #ifndef CONFIG_PARAVIRT_XXL
777 static inline void paravirt_arch_dup_mmap(struct mm_struct *oldmm,
778 struct mm_struct *mm)
783 #ifndef CONFIG_PARAVIRT
784 static inline void paravirt_arch_exit_mmap(struct mm_struct *mm)
789 #ifndef CONFIG_PARAVIRT_SPINLOCKS
790 static inline void paravirt_set_cap(void)
794 #endif /* __ASSEMBLY__ */
795 #endif /* _ASM_X86_PARAVIRT_H */